Spaces:
Runtime error
Runtime error
Group datasets by domain tab with large section headings
Browse files
app.py
CHANGED
|
@@ -14,8 +14,9 @@ from src.about import (
|
|
| 14 |
from src.display.css_html_js import custom_css
|
| 15 |
from src.display.formatting import format_timestamp_utc8
|
| 16 |
from src.utils import (
|
| 17 |
-
|
| 18 |
get_grouped_dfs,
|
|
|
|
| 19 |
prepare_ranks_table,
|
| 20 |
prepare_values_table,
|
| 21 |
)
|
|
@@ -44,23 +45,37 @@ def build_status_line(status: dict) -> str:
|
|
| 44 |
)
|
| 45 |
|
| 46 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 47 |
def refresh_leaderboard() -> tuple:
|
| 48 |
grouped = get_grouped_dfs(RESULTS_PATH)
|
| 49 |
status = load_online_status(RESULTS_PATH)
|
|
|
|
|
|
|
| 50 |
outputs: list = [
|
| 51 |
prepare_values_table(grouped["overall_values"]),
|
| 52 |
prepare_ranks_table(grouped["overall_ranks"]),
|
| 53 |
]
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
outputs.append(prepare_ranks_table(grouped["dataset_ranks"][dataset]))
|
| 58 |
outputs.append(build_status_line(status))
|
| 59 |
return tuple(outputs)
|
| 60 |
|
| 61 |
|
| 62 |
grouped_initial = get_grouped_dfs(RESULTS_PATH)
|
| 63 |
DATASETS: list[str] = grouped_initial["datasets"]
|
|
|
|
| 64 |
INITIAL_OUTPUTS = refresh_leaderboard()
|
| 65 |
|
| 66 |
demo = gr.Blocks(css=custom_css)
|
|
@@ -69,8 +84,9 @@ with demo:
|
|
| 69 |
gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
|
| 70 |
status = gr.Markdown(INITIAL_OUTPUTS[-1], elem_classes="markdown-text")
|
| 71 |
|
| 72 |
-
|
| 73 |
-
|
|
|
|
| 74 |
|
| 75 |
with gr.Tabs(elem_classes="tab-buttons"):
|
| 76 |
with gr.TabItem("Overall"):
|
|
@@ -91,25 +107,38 @@ with demo:
|
|
| 91 |
label="Ranks (average per-dataset rank, lower is better)",
|
| 92 |
)
|
| 93 |
|
| 94 |
-
for
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
|
| 98 |
-
|
| 99 |
-
value=INITIAL_OUTPUTS[2 + i],
|
| 100 |
-
interactive=False,
|
| 101 |
-
wrap=True,
|
| 102 |
-
label=f"{tab_title} · metric values (absolute, lower is better)",
|
| 103 |
-
)
|
| 104 |
)
|
| 105 |
-
|
| 106 |
-
gr.
|
| 107 |
-
|
| 108 |
-
|
| 109 |
-
wrap=True,
|
| 110 |
-
label=f"{tab_title} · ranks (lower is better)",
|
| 111 |
)
|
| 112 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 113 |
|
| 114 |
with gr.Accordion("About", open=False):
|
| 115 |
gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")
|
|
@@ -125,8 +154,8 @@ with demo:
|
|
| 125 |
refresh_outputs = [
|
| 126 |
overall_values_df,
|
| 127 |
overall_ranks_df,
|
| 128 |
-
*
|
| 129 |
-
*
|
| 130 |
status,
|
| 131 |
]
|
| 132 |
|
|
|
|
| 14 |
from src.display.css_html_js import custom_css
|
| 15 |
from src.display.formatting import format_timestamp_utc8
|
| 16 |
from src.utils import (
|
| 17 |
+
dataset_section_title,
|
| 18 |
get_grouped_dfs,
|
| 19 |
+
group_datasets_by_domain,
|
| 20 |
prepare_ranks_table,
|
| 21 |
prepare_values_table,
|
| 22 |
)
|
|
|
|
| 45 |
)
|
| 46 |
|
| 47 |
|
| 48 |
+
def _dataset_tables(grouped: dict, datasets: list[str]) -> tuple[list, list]:
|
| 49 |
+
values_tables = [
|
| 50 |
+
prepare_values_table(grouped["dataset_values"].get(dataset, None))
|
| 51 |
+
for dataset in datasets
|
| 52 |
+
]
|
| 53 |
+
rank_tables = [
|
| 54 |
+
prepare_ranks_table(grouped["dataset_ranks"].get(dataset, None))
|
| 55 |
+
for dataset in datasets
|
| 56 |
+
]
|
| 57 |
+
return values_tables, rank_tables
|
| 58 |
+
|
| 59 |
+
|
| 60 |
def refresh_leaderboard() -> tuple:
|
| 61 |
grouped = get_grouped_dfs(RESULTS_PATH)
|
| 62 |
status = load_online_status(RESULTS_PATH)
|
| 63 |
+
datasets = grouped["datasets"]
|
| 64 |
+
|
| 65 |
outputs: list = [
|
| 66 |
prepare_values_table(grouped["overall_values"]),
|
| 67 |
prepare_ranks_table(grouped["overall_ranks"]),
|
| 68 |
]
|
| 69 |
+
value_tables, rank_tables = _dataset_tables(grouped, datasets)
|
| 70 |
+
outputs.extend(value_tables)
|
| 71 |
+
outputs.extend(rank_tables)
|
|
|
|
| 72 |
outputs.append(build_status_line(status))
|
| 73 |
return tuple(outputs)
|
| 74 |
|
| 75 |
|
| 76 |
grouped_initial = get_grouped_dfs(RESULTS_PATH)
|
| 77 |
DATASETS: list[str] = grouped_initial["datasets"]
|
| 78 |
+
DOMAIN_GROUPS = group_datasets_by_domain(DATASETS, RESULTS_PATH)
|
| 79 |
INITIAL_OUTPUTS = refresh_leaderboard()
|
| 80 |
|
| 81 |
demo = gr.Blocks(css=custom_css)
|
|
|
|
| 84 |
gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
|
| 85 |
status = gr.Markdown(INITIAL_OUTPUTS[-1], elem_classes="markdown-text")
|
| 86 |
|
| 87 |
+
dataset_value_dfs: list[gr.Dataframe] = []
|
| 88 |
+
dataset_rank_dfs: list[gr.Dataframe] = []
|
| 89 |
+
dataset_idx = 0
|
| 90 |
|
| 91 |
with gr.Tabs(elem_classes="tab-buttons"):
|
| 92 |
with gr.TabItem("Overall"):
|
|
|
|
| 107 |
label="Ranks (average per-dataset rank, lower is better)",
|
| 108 |
)
|
| 109 |
|
| 110 |
+
for domain, domain_datasets in DOMAIN_GROUPS.items():
|
| 111 |
+
with gr.TabItem(domain):
|
| 112 |
+
gr.Markdown(
|
| 113 |
+
f"**{domain}** — one table per dataset (absolute metrics, lower is better).",
|
| 114 |
+
elem_classes="markdown-text",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 115 |
)
|
| 116 |
+
for dataset in domain_datasets:
|
| 117 |
+
gr.Markdown(
|
| 118 |
+
dataset_section_title(dataset, RESULTS_PATH),
|
| 119 |
+
elem_classes="dataset-heading",
|
|
|
|
|
|
|
| 120 |
)
|
| 121 |
+
dataset_value_dfs.append(
|
| 122 |
+
gr.Dataframe(
|
| 123 |
+
value=INITIAL_OUTPUTS[2 + dataset_idx],
|
| 124 |
+
interactive=False,
|
| 125 |
+
wrap=True,
|
| 126 |
+
show_label=False,
|
| 127 |
+
)
|
| 128 |
+
)
|
| 129 |
+
gr.Markdown(
|
| 130 |
+
dataset_section_title(dataset, RESULTS_PATH, ranks=True),
|
| 131 |
+
elem_classes="dataset-heading",
|
| 132 |
+
)
|
| 133 |
+
dataset_rank_dfs.append(
|
| 134 |
+
gr.Dataframe(
|
| 135 |
+
value=INITIAL_OUTPUTS[2 + len(DATASETS) + dataset_idx],
|
| 136 |
+
interactive=False,
|
| 137 |
+
wrap=True,
|
| 138 |
+
show_label=False,
|
| 139 |
+
)
|
| 140 |
+
)
|
| 141 |
+
dataset_idx += 1
|
| 142 |
|
| 143 |
with gr.Accordion("About", open=False):
|
| 144 |
gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")
|
|
|
|
| 154 |
refresh_outputs = [
|
| 155 |
overall_values_df,
|
| 156 |
overall_ranks_df,
|
| 157 |
+
*dataset_value_dfs,
|
| 158 |
+
*dataset_rank_dfs,
|
| 159 |
status,
|
| 160 |
]
|
| 161 |
|